Event detection on far edge mobile devices using delayed positioning data

Abstract

Real-time event detection is performed on nodes in an environment using position data that is not available to a node in real time but is delayed. A node performs real time event detection by predicting a position of the node based at least in part on delayed position data. The delayed position data is aligned to other sensor data. Aligning the position data may include predicting a position based on dead reckoning and/or a machine learning model. One or more collections of data, each collection including sensor data and predicted position data, is input to a model that performs event detection.

What is claimed is: 
     
       1. A method comprising:
 generating sensor data at a node in an environment, the sensor data including at least inertial data; 
 receiving delayed position data from a near edge node, wherein the near edge node generates the delayed position data from a raw position data; 
 determining a predicted position of the node based on the sensor data and the delayed position data using a position engine, wherein the position engine determines the predicted position using a dead reckoning method when a ground truth position is known and an elapsed time from moving from the ground truth position is less than a threshold value and/or a position model configured to predict the predicted position; 
 generating aligned data that includes the predicted position and the sensor data; and 
 detecting an event based on the aligned data with an event detection model. 
 
     
     
       2. The method of  claim 1 , wherein the position engine relies on the model when the elapsed time is greater than the threshold value, wherein the position model is trained using historical inertial measurements, bearing and rich movement data, and/or classifications of an operational state. 
     
     
       3. The method of  claim 2 , wherein the position model is trained using historical position data including historical trajectories. 
     
     
       4. The method of  claim 3 , wherein training using the historical position data includes a supervised learning task using past trajectories to determine a next position based at least on one or more previous positions. 
     
     
       5. The method of  claim 2 , wherein determining the predicted position includes interpolating a first delayed position and a second delayed position relative to a sensor data collection. 
     
     
       6. The method of  claim 2 , wherein determining the predicted position includes predicting a position for a most recent sensor data collection. 
     
     
       7. The method of  claim 2 , wherein determining the predicted position includes predicting a position for a time after the most recent sensor data collection. 
     
     
       8. The method of  claim 7 , wherein a sensor collection for the time after the most recent sensor data collection is generated by extrapolating the most recent sensor data collection or by copying the most recent data collection into the sensor collection for the time after the most recent sensor data collection. 
     
     
       9. A non-transitory storage medium having stored therein instructions that are executable by one or more hardware processors to perform operations comprising:
 generating sensor data at a node in an environment, the sensor data including at least inertial data; 
 receiving delayed position data from a near edge node, wherein the near edge node generates the delayed position data from a raw position data; 
 determining a predicted position of the node based on the sensor data and the delayed position data using a position engine, wherein the position engine determines the predicted position using a dead reckoning method when a ground truth position is known and an elapsed time from moving from the ground truth position is less than a threshold value and/or a position model configured to predict the predicted position; 
 generating aligned data that includes the predicted position and the sensor data; and 
 detecting an event based on the aligned data with an event detection model. 
 
     
     
       10. The non-transitory storage medium of  claim 9 , wherein the position engine relies on the model when the elapsed time is greater than the threshold value, wherein the position model is trained using historical inertial measurements, bearing and rich movement data, and/or classifications of an operational state. 
     
     
       11. The non-transitory storage medium of  claim 10 , wherein the position model is trained using historical position data including historical trajectories. 
     
     
       12. The non-transitory storage medium of  claim 11 , wherein training using the historical position data includes a supervised learning task using past trajectories to determine a next position based at least on one or more previous positions. 
     
     
       13. The non-transitory storage medium of  claim 10 , wherein determining the predicted position includes interpolating a first delayed position and a second delayed position relative to a sensor data collection. 
     
     
       14. The non-transitory storage medium of  claim 10 , wherein determining the predicted position includes predicting a position for a most recent sensor data collection. 
     
     
       15. The non-transitory storage medium of  claim 10 , wherein determining the predicted position includes predicting a position for a time after the most recent sensor data collection. 
     
     
       16. The non-transitory storage medium of  claim 15 , wherein a sensor collection for the time after the most recent sensor data collection is generated by extrapolating the most recent sensor data collection or by copying the most recent data collection into the sensor collection for the time after the most recent sensor data collection.